Hype Index: 'Your Electricity Bills Have Gone Up by as Much as 267%'

TL;DR: Data centers are raising electricity bills in PJM territory, and the number everyone quotes is still wrong. The viral "267%" is a wholesale nodal price change from April 2020 to April 2025, not a residential bill, and wholesale supply is only 30% to 50% of what you pay; PolitiFact rated it Mostly False on 2026-06-12. I pulled EIA's actual retail table (5.6.A, May 2026 vs May 2025) on 2026-08-15: the median PJM-state residential price rose 10.26% in twelve months against 3.96% everywhere else, a 6.3-point gap that widens to 9.96 points in the eight jurisdictions wholly inside PJM. That gap is suggestive, not proof. Score: 45% noise / 55% signal.

What is the claim?

Senator Elizabeth Warren, in a June 5, 2026 post on X:

If you live near one of these large data centers, your electricity bills over the last five years have gone up by as much as 267%.

PolitiFact rated it Mostly False on June 12, 2026. The reason is a category error, not a factual one. The 267% is real, and it traces to a September 2025 Bloomberg analysis of wholesale prices at individual grid nodes near data centers between April 2020 and April 2025. It is a wholesale nodal figure. It is not anyone's bill.

Yale economist Kenneth Gillingham gave PolitiFact the sentence that settles it: "the wholesale nodal electricity prices only raise the 'supply' component of electricity bills." That supply component is roughly 30% to 50% of a residential bill. The rest is transmission, distribution and taxes. Warren's own office pointed PolitiFact to CBS News and Fortune coverage that had made the same substitution first, which is how a wholesale number becomes a household number: nobody re-reads the axis label.

What do the retail numbers actually show?

Nobody publishes the comparison I wanted, so I ran it. On 2026-08-15 I downloaded EIA's Electric Power Monthly Table 5.6.A, the average retail price to residential customers by state, which at that snapshot carried May 2026 against May 2025. I parsed all 51 jurisdictions and split them into PJM territory (the 13 states plus DC that PJM serves in whole or part) and everywhere else.

Group	n	Median 12-month change
PJM territory	14	+10.26%
Everywhere else	37	+3.96%
PJM, restricted to jurisdictions essentially wholly inside it	8	+13.92%
Texas (ERCOT)	1	+5.86%
All 51	51	+5.39%

The PJM median runs 6.30 points above the rest of the country. Restrict to the eight jurisdictions that are effectively all PJM (DC, Delaware, Maryland, New Jersey, Ohio, Pennsylvania, Virginia, West Virginia) and the gap widens to 9.96 points, which is what you would expect if the effect is real: the cleaner the exposure, the bigger the number.

Individual states: Illinois +28.36%, DC +24.33%, Virginia +15.40%, Maryland +14.58%, Ohio +14.22%, New Jersey +13.62%, Pennsylvania +11.72%. Texas, which absorbed comparable data-center growth under a different market design, rose 5.86%.

EIA revises these monthly. That table is a snapshot taken on 2026-08-15 and it will not match the same URL in six months.

Does that gap prove data centers did it?

No, and this is where the topic eats people.

State retail electricity prices move on gas prices, storm-cost recovery, transmission build, rate-case timing and the political calendar of fifty separate utility commissions. A 6-point median gap between two groups of states is consistent with the data-center story and also consistent with several others. I did not control for anything, because with 51 observations and a dozen plausible confounders there is nothing honest to control for.

My own data contains the counterexamples. Hawaii, which has no meaningful data-center load, rose 26.7%, second-highest in the country. Connecticut fell 13.4%. West Virginia sits inside PJM and rose 3.83%, below the national median. Illinois, my single largest increase, is only partly inside PJM and had a rate case land in the window. If the PJM grouping were capturing a clean causal signal, West Virginia and Illinois would not both be in it.

Treat the gap as what it is: a consistent, checkable pattern that survives an obvious test, pointing the same direction as the mechanism evidence. That is worth something. It is not attribution.

What is the mechanism, and who names it?

The specific machinery is PJM's capacity auction, where generators are paid to be available. It cleared at $28.92 per megawatt-day for 2024/25 and $269.92 for 2025/26, roughly a nine-fold jump, and that flows into retail bills across the footprint whether or not a data center is anywhere near you.

SemiAnalysis put numbers on the household end: PJM residents paying "$25-30 more per month than two years ago," about 15%, driven mainly by the capacity mechanism rather than data-center consumption itself, against ERCOT power futures that "moved only a few percent in the past year." Their conclusion is "empirically the fault is government policy, not AI." Read that with the conflict in view: SemiAnalysis is a paid research shop selling analysis to an industry with a direct commercial interest in the buildout not being blamed. Their arithmetic is checkable and their framing is not neutral.

The other main attribution source has the mirror-image problem. PJM's Independent Market Monitor found data-center load growth was "the primary reason" for recent capacity conditions and put the effect at $9.3 billion, or 174%, for the 2025/26 delivery year. That is a party's estimate inside a contested proceeding, not neutral arithmetic, and it should be cited with the same skepticism as the vendor number.

So I counted who tells readers about the mechanism at all. I fetched and classified six general-audience articles blaming data centers for bills. Two distinguish the capacity market from data-center consumption: PolitiFact, and the New Jersey State Policy Lab, which separates a barely-detectable local effect from a real regional one. Four do not: NC State models dispatch prices, EESI points at utility commissions approving rate increases, Consumer Reports offers "utilities are building infrastructure, and then we all pay for it," and Forbes covers tariff design without naming the auction.

Those four are not wrong. Rate-base recovery and tariff design are genuine channels. But a reader who never learns that a capacity auction exists cannot evaluate any proposed fix, because most of the proposed fixes are auction reforms.

What should you do instead?

Check your own bill's supply line against your utility's rate history rather than a national percentage, since the supply component is the only part wholesale prices touch directly. If you want to know whether your state is exposed, the question is not how many data centers are nearby, it is whether you are inside PJM and when your utility's next rate case is, and not whether the projects near you get built, since the widely reported "half of them are canceled" figure was 30 to 50% delayed, globally. And when a number like 267% arrives, ask which price it measures before asking who to be angry at, because the same buildout that may or may not be carrying US GDP growth produces exactly this genre of confidently mislabeled statistic. It is the same failure mode that makes the bubble question unanswerable in public and the same one that makes AI cost claims so hard to pin down.

Verdict

The direction is right and the famous number is not. Data centers are a real driver of a real capacity-price spike that is really on PJM bills, and the retail data I pulled is consistent with that at a 6.3-point median gap. But 267% is a wholesale nodal price masquerading as a household bill, the honest household figure is closer to $25 to $30 a month, and the mechanism is an auction design most coverage never mentions. The claim is a true thing wearing a false number.

Verdict: 45% noise / 55% signal. The buildout is moving your bill. It is not moving it by 267%.